tgt
stringlengths
1
2.76k
src
stringlengths
1
3.03k
lp
stringclasses
15 values
[PERSON8] Ale přesto, jak sleduji soubory, který je potřeba zkonvertovat ručně, abych mohl napsat skript, a ten by se mohl postarat alespoň o ty známý, a samozřejmě, pokud přidáme nějaký nový soubory, můžu se na to podívat později.
(PERSON8) But still, if I am keeping track of the files that need to be manually converted, so that I can write a script, and it could take care of the known ones at least, and of course, if we add some new files, I can see for that later.
en-cs
[PERSON6] Takže se jedná o důležitou věc, na kterou jsme měli narazit už před několika měsíci.
(PERSON6) So this is an important thing that we should have come across months ago.
en-cs
Nezapomeňte, že už od srpna vám říkám, že chceme vyhodnocovat na pozdějším testovacím souboru.
Remember, it is since August that I've been telling you that we want to evaluate on a later test set.
en-cs
A pokaždé jste všichni...
And every time, you all...
en-cs
Nejde jen o vás osobně, [PERSON8].
It is not just you personally, [PERSON8].
en-cs
Byly tam samozřejmě i další věci.
There were other things, obviously.
en-cs
Ale pokaždé, když jste o tom přemýšleli, říkali jste si „to bude snadný, prostě to spustíme“.
But every time you thought about that, you thought "this will be easy, we will just run it".
en-cs
A teprve až to opravdu uděláte, uvidíte všechny ty problémy.
And only when you actually do it, do you see all the problems.
en-cs
Tak je to známý, běžný.
So, this is known, common.
en-cs
A je to tak vždycky.
It is always like that.
en-cs
Takže tohle je vzkaz pro všechny.
So that's a message for everybody.
en-cs
Než se o to pokusíte, nikdy nevíte, v čem bude problém.
Before you try, you never know what the problem will be.
en-cs
Tak například hovor, který jsem měl, kdy se mnou někdo dělal rozhovor, a tak si chtěl nahrát video.
So, for example, the call that I had, in which someone was interviewing me, so they wanted to record the video.
en-cs
Zoom mi nikdy nefungoval tak špatně jako dneska při nahrávání.
Zoom never worked as badly for me as it did today for the recording.
en-cs
Než to uděláte, nemůžete říct, zda to bude fungovat, nebo ne.
So, before you do it, you cannot say whether it will work or not.
en-cs
A tady narážíme na problémy s konverzí formátů souborů.
And here we are running into the problems of conversion of the file formats.
en-cs
Doufejme tedy, že se nám podaří rychle získat čísla.
So, hopefully, we will be able to quickly get the numbers.
en-cs
Máte zatím nějaké údaje o chybovosti slov?
So, do you have any word error rates so far?
en-cs
[PERSON8] Ne, nezpracoval jsem je.
(PERSON8) No, I have not processed it.
en-cs
Takže [PERSON1] řekl, že jakmile dostanou přepis, budeme to dělat společně.
So [PERSON1] said that, once, they get the transcript, we will do it together.
en-cs
A pak souhlasím s tím, že v tomhle slovníku je spousta šumu.
And then, I agree that there is a lot of noise in this dictionary.
en-cs
Tak jsem vás požádal, abyste měli různý verze tohohle slovníku, a to tak, že vezmete pouze ty slova, který byly pozorovaný, pětkrát.
So, I asked that you would have various versions of this dictionary by taking only the words that were observed, five times.
en-cs
A pouze ty výslovnosti, který byly pozorovaný třikrát nebo vícekrát, nebo něco takového.
And only the pronunciations that were observed three times or more, or something like that.
en-cs
Takže takhle se tyhle náhodný chyby, jako je příklad s těmahle čtyřma IBM místo IBM.
So, that way, these random errors, such as the example with this four IBM instead of IBM.
en-cs
Tohle nebude tak častý.
This will not be so frequent.
en-cs
Ale přesto budete mít varianty ve výslovnosti, protože pokud osoba, řekne to „alzo“, pokud si pamatujete, pokud řekne to „alzo“ se „Z“ za, za rozhovor 20krát, pak ho uvidíte v datech se „Z“.
But still, you will have the variants in pronunciation, because if the person, says this "alzo", if you remember, if he says this "alzo" with a "Z" 20 times in the in the talk, then you will see that in the data with the "Z".
en-cs
A pokud se mu to někdy podaří říct i správně, pak jo, zas to uvidíte v datech.
And if he sometimes manages to say also correctly, then yeah, again, you will see that in the data.
en-cs
Takže bych chtěl, aby [PERSON4] byl v kontaktu s [PERSON2].
So, I would like, [PERSON4] to be in touch with [PERSON2].
en-cs
Možná budete pokračovat v hovoru, co?
Maybe you continue on the call, right?
en-cs
Budu už muset odejít, protože musím dát dětem oběd a tak vůbec.
I will have to leave now, because I need to give the kids lunch and so on.
en-cs
Ale prosím, zůstaňte u toho hovoru a společně zjistěte, jak ten slovník používat.
But please stay on this call and figure out jointly how to use that dictionary.
en-cs
Takže, [PERSON4], prosím, ukaž [PERSON2], jaký slovník systém přijímá.
So [PERSON4], please show [PERSON2] what the dictionary is that the system is accepting.
en-cs
Sdílej obrazovku a ukaž ji.
Share your screen and show it.
en-cs
A [PERSON2], podívej se prosím na to a pomoz s konverzí slovníku, který jste na ten soubor pustili.
And [PERSON2], please, have a look at that and help with converting the dictionary that you emitted to that file.
en-cs
Jedna z věcí, která bude pořád ještě potřeba, je náhrada jazykového modelu.
One thing that will still be needed is the language model substitute.
en-cs
Ale všechny ty slova by měly být známý.
But all these words should be known words.
en-cs
Takže ,[PERSON4], musíš zjistit, jaký slova jsou v jazykovém modelu.
So [PERSON4], you need to figure out what words are in the language model.
en-cs
A slova, která jsou v jazykovém modelu, by se měly jednoduše zkopírovat.
And the words which are in the language model should be simply copied.
en-cs
Takže slovník bude mít tři sloupce.
So, the dictionary will have three col ns.
en-cs
Grafém tak, jak je na výstupu.
The grapheme, as it is output.
en-cs
Jeden rozpoznaný.
One recognized.
en-cs
Fonémy tak, jak je [PERSON2] navrhl.
The phonemes, as [PERSON2] proposed.
en-cs
A bude existovat víc řádků s různými variantami fonémů.
And there will be multiple lines with different phoneme variations.
en-cs
Třetí sloupec bude zase stejný pro všechny z nich a zase bude mít stejnou formu grafému.
And the third col n will be also the same again for all of those, and again there will be the same grapheme form.
en-cs
A to je ta chvíle, kdy to jazykový model vidí.
So that is when the language model sees.
en-cs
A myslím, že tímhle způsobem by tyhle systémy měly být schopný ho načíst.
And I think that this way, these systems should be able to load it.
en-cs
A možná narazíte ještě na jeden další problém: že není připravený na dva velký vlastní slovníky.
And you will also possibly run into one more problem: that it is not ready for two big custom dictionaries.
en-cs
Takže i to je něco, co je potřeba otestovat, ale prosím, otestujte to vy dva společně.
So, this is also something that has to be tested, but please test it, the two of you together.
en-cs
Takže [PERSON2] ví, co ve slovníku vytvořil.
So [PERSON2] knows what he created in the dictionary.
en-cs
A ty víš, jak slovník vypadá, když ho vytváříš ručně, a tyhle dvě znalosti musíte spojit dohromady.
And you know what the dictionary looks like when you are creating it manually, and you need to put these two pieces of knowledge together.
en-cs
Aha, takže to funguje s tím, s tím vygenerovaným slovníkem.
Ah, so that it works with the, with the generated dictionary.
en-cs
[PERSON4] Dobře.
(PERSON4) Okay.
en-cs
[PERSON6] Takže, [PERSON2], dává to smysl?
(PERSON6) So [PERSON2], does this make sense?
en-cs
[PERSON2] Jo, jistě, probereme to.
(PERSON2) Yeah, sure, we'll discuss it.
en-cs
[PERSON6] Jo, takže je ještě něco, [PERSON2], co máš?
(PERSON6) Yeah, so is there anything else, [PERSON2], that you have?
en-cs
[PERSON2] No, možná zajímavá informace pro Dominika.
(PERSON2) Well, maybe interesting information for Dominik.
en-cs
Dokončuji výcvik německého ASR, který se dá použít pro časovou značku.
I'm finishing training of a German ASR that may be used for time stamping.
en-cs
Nejsem si jistý, jak dobrý to bude, protože se to snažím dělat na (Libry) Speech.
I'm not sure how good it will be because I'm trying to do it on (Libry) Speech.
en-cs
Když jsem to stahoval, tvrdili, že má přes 600 hodin.
And, though, when I was downloading it, they claimed it had more than 600 hours.
en-cs
Ale pak se při trénování ukázalo, že skutečný tréninkový soubor obsahuje jenom asi 300 hodin.
But then the training said that the actual training set contained only around 300 hours.
en-cs
A stále si nejsem jistý, zda těch 300 hodin neobsahuje podobný věty.
And I'm still not sure whether these 300 hours do not contain similar sentences.
en-cs
Nebo vlastně tytéž věty, ale pronášený různými mluvčími.
Or actually the same sentences but spoken by different speakers.
en-cs
Například v angličtině nebo češtině jsem dřív pozoroval prudší konverze.
For example, in English or Czech I used to observe steeper conversions.
en-cs
A teď už nekonvertuje tak rychle.
And now it won't convert so fast.
en-cs
Takže pokud během trénování vezmeme několik vzorků.
So, if we take some samples during the training.
en-cs
Pak se ve výstupu ASR pořád vyskytují určitý závažný chyby.
Then there are still some serious errors in the ASR output.
en-cs
Doufám tedy, že na časovou značku je to dostatečně dobrý.
So, I'm hopeful that for the time stamping, it is good enough.
en-cs
Nebo to aspoň můžeme zkusit.
Or at least we might try it.
en-cs
[PERSON17] Dobře, to je dobrý, jo.
(PERSON17) Okay, that's good, yeah.
en-cs
Tak jsem ti začal psát e-mail, ale myslím, že jsem ho nikdy nedokončil, protože jsi mi připomněl, [PERSON8], že filtrování nadávek ještě není integrovaný.
So, I started, but I think I never finished an email to you, because you reminded me, [PERSON8], that your profanity filtering is not yet integrated.
en-cs
A myslím, že je to taky důležitá zpráva pro [PERSON5], který se z hovoru zase vytratil.
And I think this is also an important message for [PERSON5], who has disappeared again from the call.
en-cs
Důležitý poselství tedy je, že ano, je velmi dobrý, že aktivně usilujete o to, aby se vaše výsledky integrovaly, a měl by to tak dělat každý.
So, the important message is that yes, it's very good that you are actively pushing so that your results are integrated, and everybody should do so.
en-cs
A zároveň potřebujeme mít nastavení, abyste ho mohli sami integrovat a testovat.
And at the same time, we need to have the setup, so that you can actually integrate and test it yourself.
en-cs
Říkám tomu „integrace svépomocí“.
So, I call it "do-it-yourself integration".
en-cs
Takže, [PERSON8], když pracuješ s [PERSON5] a dokumentujete, jaká jsou nastavení, ujistěte se, že je to dostatečně dobře otestovaný kolegy, například [PERSON4] nebo dokonce [PERSON9] z hlediska kontrol jazykového modelu a všema ostatníma.
So, [PERSON8], when working with [PERSON5] and when documenting what the setups are, make sure it is tested well enough by colleagues, such as [PERSON4] or even [PERSON9] for the language model checks and everybody else.
en-cs
Takže kdykoli někdo vyvine novou užitečnou komponentu, měl by mít dostatečně snadno k dispozici celou pajplajnu, aby ji mohl sám otestovat.
So, whenever someone develops a new useful component, the full pipeline should be accessible to him reasonably easily, so he can test it himself.
en-cs
Tato integrace svépomocí je tedy důležitá, protože jinak všechno zůstane na tobě, [PERSON8], a ty se nechceš nechat přetěžovat.
So, this do-it-yourself integration is important, because otherwise it will all remain on you, [PERSON8], and you don't want to be overloaded.
en-cs
Proto chceš těmhle lidem poskytnout vstupy a výstupy jako první testovací přístup, který už proběhl.
So, you want to provide these people with inputs and outputs as the first testing approach, which has already been done.
en-cs
Je to tak, [PERSON4]?
[PERSON4], right?
en-cs
Filtrování nadávek, bylo už otestovaný na protokolech?
The profanity filtering, has it been tested on logs?
en-cs
Myslím, že bylo.
I think it was.
en-cs
[PERSON8] Ano, myslím, že bylo testované na protokolech.
(PERSON8) Yeah, it was tested on logs, I think.
en-cs
Takže teď nastal čas otestovat ho na ostrých pajplajnách.
So, now it's time to test it on the live pipelines.
en-cs
A opět si myslím, že je lepší, když [PERSON8] vysvětlí [PERSON4], jak to má udělat, aby si [PERSON4] spustil sám pro sebe některý převaděče a živě přehrál některý z problematických souborů, třeba za pomoci [PROJECT8] nebo jakkoli jinak.
And again, I think it's better if [PERSON8] explains to [PERSON4] how to do it, so that [PERSON4] runs it for himself some of the workers and live playing some of the problematic file like using [PROJECT8] or whatever.
en-cs
Jednoduše je přehraj.
Simply play them.
en-cs
Sledujte zvukový výstup na svém počítači a podívejte se, jak to funguje.
Follow the sound output on your machine and see how that works.
en-cs
Protože teprve při skutečným nastavení se objeví skutečné chyby.
Because only when doing the real setup will the true errors appear.
en-cs
Je důležitý ho nejprve odladit pomocí souborů protokolu a pak je důležitý ho odladit v pajplajně.
It is important to first debug it, using the log files, and then it's important to debug it in the pipeline.
en-cs
A pokud by to ladění mohl provádět autor dané komponenty, v tomhle případě [PERSON4], bylo by to pro nás všechny nejefektivnější.
And if this debugging can be done by the author of that component, here, in this case [PERSON4], it would be most efficient for all of us.
en-cs
Takže, [PERSON8], prosím, potvrď, že s touhle myšlenkou jakoby integrace svépomocí souhlasíš.
So, [PERSON8], please confirm that you agree with this idea of like do-it-yourself integration.
en-cs
[PERSON17] Jo, takže pro angličtinu s více přízvuky.
(PERSON17) Yeah, so for the multi-accent English.
en-cs
Takže to teď [PERSON14] dáváme dohromady do jednoho technického řešení.
So, we are now [PERSON14] putting that together to just one technical solution.
en-cs
Současný nápad, na kterém [PERSON14] pracuje, spočívá v tom, že bude vytvářet nový věty se slovy, který byl řečený v jiných větách, a provede to napříč různými mluvčími.
The current idea that [PERSON14] is working on is that he will create new sentences with words that were spoken in other sentences, and he will do this across different speakers.
en-cs
Bude se tedy jednat o věty s více mluvčími, a proto by se mohla zlepšit i odolnost vůči různým přízvukům těchhle mluvčích.
So, it will be really multi-speaker sentences, and therefore the robustness to the different accents of these speakers could be also improved.
en-cs
Aha, takže to je jeden částicový experiment.
Ah, so that's one particle experiment.
en-cs
A později možná uděláme něco víc s tou záležitostí s více přízvuky.
And later on, we may do something more about the multi-accent thing.
en-cs
Takže tyhle nový věty se vlastně snaží vyřešit dva problémy jedním experimentem.
So, these new sentences will actually try to solve two problems with one experiment.
en-cs
Jedním z problémů je implicitní jazykový model.
One problem is the implicit language model.
en-cs
Takže systém ASR musí vidět co největší soubor vět.
So, the ASR system has to see the largest possible set of sentences.
en-cs
A my budeme vytvářet nový věty z pouze textového jazykového modelu přidáním zvukové části, takže jazykový model bude pro ASR lepší a robustnost vůči různým mluvčím bude taky lepší.
And we are going to create new sentences from the text-only language model by adding the sound part to that, so that the language model will be better for the ASR, and the robustness to different speakers would be also better.
en-cs